Bind the user-transcribed hosted result to evaluated revision d9d611833d340c72d90a97d995a94031b798cf7c. Record scored accuracy as 985/1000 (98.5%, displayed 99%), the separate tier-0 diagnostic, per-tier counts, runtime, and artifact size. Model code, weights, and manifest are unchanged.
Remove an unsupported held-out-prime implication. Training used independently sampled random moduli; separately generated cases served as selection and qualification gates. Model code and tensors are unchanged.
Publish MiniNeuralHorner v0.2 development snapshot
126,603-parameter B127 artifact. Trained through L512, update-zero qualified at L1024, packaged for a SAIR Playground research evaluation. TrickyRex/bitserial-modmul-v8 is unchanged.
